I've been struggling with the new Opus release because it seems to completely ignore my project instructions. I'm trying to use it for creative writing, but it keeps bringing up the names "Chen", "Marcus", and "Sarah" in every chat, despite me creating a project specifically to ban those names. With the previous version, 3.7, it would generally follow my requests and wouldn't repeat mistakes after I pointed them out. But now, with Opus, it seems like no matter how many times I ask it not to use those names, it just apologizes without addressing the issue. Has anyone else experienced this problem and does anyone know how to make Opus actually follow instructions?
4 Answers
Yeah, I've found project instructions can be hit or miss. Have you tried putting your instructions right at the start of your prompts? That might help clarify your intent from the get-go.
Try giving Opus a specific list of names it can use instead of just telling it what not to do. Sometimes focusing on the positives helps more than listing restrictions.
Could you clarify your initial prompt? Sometimes the way we phrase things can change how the AI interprets instructions.
It seems like the issue is similar to telling someone not to think of a pink elephant. By mentioning forbidden names, you might actually be making them more prominent in its context. It might be better to focus your instructions on the names you want it to use instead of what you don’t want.
This is really useful info! It seems like the less you mention the unwanted names, the better.
Exactly! Instead of saying what not to do, give it good examples of what you want. This could really change how it generates names.